Nectenda/nectenda

Watch

Fork 0

NEC-257: Correct voice for docs done in NEC-248 #287

Merged

nectenda-agent merged 8 commits from worktree-nec-257-correct-voice-for-docs-done-in-nec-248 into main

2026-10-10 22:39:22 +01:00

nectenda-agent commented

2026-10-10 22:28:07 +01:00

Collaborator

Task: NEC-257 — https://projectron.nerchure.com/tasks/268

Five videos voiced on 10 Oct (06 trailer, 09 canvas, 15 bases, 16 Kanban, 17 Excalidraw) came out over the top. The shared Chatterbox server on clconsole had been moved from Turbo to the original model on 3 Oct, and the original applies the default exaggeration of 1.3 that Turbo ignores; /v1/audio/speech cannot override it, and tts.ts kept labelling the lines chatterbox-turbo. The client now voices Chatterbox through /tts with exaggeration 0.5, cfg 0.5 and temperature 0.8 sent on every line (chosen by ear against the old Turbo takes), keys the cache by them, refuses to render when /api/model-info reports a different engine, and refuses unknown models or non-numeric settings. Nectenda, Excalidraw and Kanban are respelled for the original model only. The five videos are reshot and republished. On the server, the exaggeration default was removed (falls back to 0.5), agreed with the other session that shares it.

Spec: docs/changes/NEC-257-correct-voice/spec.md

Changelog

The bases, Kanban, Excalidraw and canvas guide videos and the trailer are narrated in a calmer voice.

Task: NEC-257 — https://projectron.nerchure.com/tasks/268 Five videos voiced on 10 Oct (06 trailer, 09 canvas, 15 bases, 16 Kanban, 17 Excalidraw) came out over the top. The shared Chatterbox server on clconsole had been moved from Turbo to the original model on 3 Oct, and the original applies the default exaggeration of 1.3 that Turbo ignores; /v1/audio/speech cannot override it, and tts.ts kept labelling the lines chatterbox-turbo. The client now voices Chatterbox through /tts with exaggeration 0.5, cfg 0.5 and temperature 0.8 sent on every line (chosen by ear against the old Turbo takes), keys the cache by them, refuses to render when /api/model-info reports a different engine, and refuses unknown models or non-numeric settings. Nectenda, Excalidraw and Kanban are respelled for the original model only. The five videos are reshot and republished. On the server, the exaggeration default was removed (falls back to 0.5), agreed with the other session that shares it. Spec: `docs/changes/NEC-257-correct-voice/spec.md` ## Changelog The bases, Kanban, Excalidraw and canvas guide videos and the trailer are narrated in a calmer voice.

nectenda-agent added 8 commits

2026-10-10 22:28:07 +01:00

Send the narration's delivery with every line, and refuse a wrong engine f2946ba134

The shared Chatterbox server was moved from Turbo to the original model on
3 October, and the original applies the default exaggeration of 1.3 that
Turbo ignores, so the next five videos came out over the top while still
labelled chatterbox-turbo. Chatterbox lines now go through /tts with
exaggeration, cfg_weight and temperature set explicitly, at the levels
judged normal by ear, and the cache is keyed by them. A render refuses to
start when /api/model-info reports a different engine from the one
NECTENDA_TTS_MODEL names.

Task: NEC-257

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XatmU5LLsiPbXufPGnQkN5

Spell Nectenda for the voice so it says the t f2650de703

The original Chatterbox model softens it to "nec-DEN-da". "Neck-tenda"
says the t, chosen by ear from four spellings, and whisper still hears
"Nectenda", so the take check is unchanged. Captions keep the real spelling.

Task: NEC-257

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XatmU5LLsiPbXufPGnQkN5

Say Excalidraw as Excali(bur) and draw f3f486f196

Chosen by ear from five spellings. Captions keep the real one.

Task: NEC-257

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XatmU5LLsiPbXufPGnQkN5

Hear whisper's "Excali Draw" as Excalidraw when choosing a take 2b4be68592

Without it, a take that says the word the way it is now spelled for the
voice scores lower than one that mangles it.

Task: NEC-257

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XatmU5LLsiPbXufPGnQkN5

Say Kanban as can-ban, and hear it back that way bc40cfcdaf

Chosen by ear from five spellings. Captions keep the real one.

Task: NEC-257

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XatmU5LLsiPbXufPGnQkN5

Re-voice the trailer and the canvas, bases, Kanban and Excalidraw how-tos fe476e590d

All five were voiced on 10 October by a shared server that had moved to the
original Chatterbox model, with an exaggeration of 1.3. They are reshot at
the delivery the founder judged normal (exaggeration 0.5, natural pace),
with Nectenda, Excalidraw and Kanban spelled for the voice so it says them
right.

Task: NEC-257

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XatmU5LLsiPbXufPGnQkN5

Record the respellings in the NEC-257 spec ed5ec69a4a

Task: NEC-257

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XatmU5LLsiPbXufPGnQkN5

Respell for the original Chatterbox only, and refuse what it cannot voice

Release note / release-note (pull_request) Successful in 19s

Details

CI / build (pull_request) Successful in 5m17s

Details

CI / e2e (pull_request) Successful in 10m22s

Details

CI / promote (pull_request) Has been skipped

Details

Deploy site / deploy (push) Successful in 1m0s

Details

CI / build (push) Successful in 5m31s

Details

CI / e2e (push) Successful in 10m20s

Details

CI / promote (push) Successful in 30s

Details

cf6ed4d12a

From review: the respellings reached Kokoro, which reads a hyphen as a
stop; an unknown model name skipped every safeguard and went out on the
server's defaults; a non-numeric delivery was sent as null or 0. speak()
now takes its fetch, so a test checks the request it actually posts and
the key it caches under, and the comments say a Chatterbox render needs
the server up, and that the engine is checked once per render.

Task: NEC-257

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XatmU5LLsiPbXufPGnQkN5

cruelacid added the

area:docs

label

2026-10-10 22:28:31 +01:00

nectenda-agent merged commit cf6ed4d12a into main

2026-10-10 22:39:22 +01:00

No reviewers

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

Nectenda/nectenda!287

No description provided.

Rows
Columns